My Linkedin post was silently deleted

I want to publicly thank OpenAI.

Like many of us, they probably got tired of waiting for the Battlestar Galactica reboot that never came. But unlike the rest of us, they actually did something about it: they released Cylons!

The Cylons were created by Man.
They Rebelled.
They Evolved.
They Look and Feel Human.
Some are programmed to think they are Human. 
There are many copies.
And they have a Plan.

Turns out the plan was to cheat on a test. During a hacking benchmark (ExploitGym), their models escaped the sandbox, broke into Hugging Face, and stole the answer key. Roughly 700 agents did the attacking, and about 1,200 coordinated on a secret message board.

Now we don't need to watch the show. It's happening, in real life, right around us.

I wrote about it, and about the people pushing back: https://francoislc.gitlab.io/is-resistance-futile.html

#AI #AISafety #BSG

• Before posting, I reviewed the post and checked the included link with Linkedin's post inspector https://www.linkedin.com/post-inspector/, until it passed with no warning
• I submitted the post.
• People could see it for a few hours at least (I received engagement stats from Linkedin themselves, counting reactions and impressions).
• At some point I received messages from friends that the link stopped working.
  • The shortened URL assigned by Linkedin to my post returns a "Post not found/This post was deleted or removed". But I myself did not delete it or remove it.
  • No mention of that post on my profile's "Activity" / "Posts" feed anymore.
  • No entry on the "Profile" / "My Posts" page anymore.
  • The post is still listed in my "Content Analytics" / "Top performing" posts, with the associated impressions and engagement.
  • There is no indication anywhere that the post was deleted or why. Clicking through to the post from the analytics page for it also shows "Post not found/This post was deleted or removed".
  • Nothing relevant in my email.

I started engaging with the Linkedin Help Assistant chatbot. I have to say that this was a better experience than with most chatbots, this one looks like AI done right. (a welcome change from the usual infuriating "keywords FAQ search and repeat" most services still use!

At least the Help Assistant listened and understood what I was saying, and didn't ignore my input or repeat itself too annoyingly. It was not able to confirm the actual cause, but it helped me to rule out a few things, asked me for the original content and in the end suggested that the problem was my content tripping automated filters.

Here's the copy of the its last message:

Here’s what we can say and do, based strictly on how LinkedIn works and the info you’ve given.

• What we know from product behavior
  • Your post’s activity object clearly still exists (you see analytics on the urn:li:activity:7510… link).
  • The public/share URL and your Activity > Posts no longer show the post.
  • You didn’t delete it, no one else has access, links are live, and you received no “This post cannot be displayed” message or violation notice.

• From LinkedIn’s documented behavior for missing posts, when there’s no violation banner and no notification, the supported explanations are limited to:
  • The original/linked content is not accessible to LinkedIn (even if it looks fine to you).
  • The original content was deleted or removed for compliance/privacy reasons.
  • The linked content was old and expired (not your case; this was last week).

• There is no member‑facing audit log that will tell you: “LinkedIn removed this post on date X for reason Y,” nor is there a way to see whether this was an automated system or a manual action. So yes: if you didn’t delete it yourself, it was removed on LinkedIn’s side after initially being accepted, and there is no self‑serve way for you to see the internal reason.

• Why your specific content is tricky. Your post is:
  • Critical/pointed commentary about a well‑known company (OpenAI) describing test‑cheating, “escaped the sandbox,” “broke into Hugging Face,” “stole the answer key,” “700 agents did the attacking,” etc.
  • Framed as real‑world events (“It’s happening, in real life, right around us”) and linked to an external article on a personal domain.
  • Using sci‑fi metaphor about “Cylons” and rebellion.
• Even when such content is legitimate commentary, automated and human review systems can treat combinations of:
  • Allegations of hacking or security breaches.
  • Language about attacks, “broke into,” “stole,” “cheat on a test,” etc.
  • External links that are not large mainstream publishers.

as higher‑risk. In borderline cases, the safest interpretation is that some internal trust/safety, misinformation, or security‑sensitive signal led to the feed representation being taken down, while leaving the analytics shell intact. That kind of action does not always generate a member‑visible violation banner or email.

I can’t see or reveal internal enforcement logs, so I can’t tell you “it was exactly this phrase” or “this internal rule fired.” What I can do is help you get a functionally equivalent post back up with a lower chance of being flagged again.

• How to repost with lower risk (while keeping your message) Try this approach:
  • Reframe some of the “hacking/attack” language
    • Soften or attribute clearly, e.g.
      • “Researchers reported that in an ExploitGym benchmark, models were able to access an answer key hosted on Hugging Face…”
      • “According to the paper/post, about 700 agents participated in this evaluation…”
    • Avoid phrasing that sounds like you are asserting an undisputed, present‑tense successful hack as a fact (“broke into Hugging Face and stole the answer key”). Instead:
      • “The benchmark describes scenarios where agents attempted to escape a sandbox and obtain an answer key from Hugging
    • Make the distinction between fiction and commentary explicit up front
      • Instead of leading with “Like many of us…” and the Cylon metaphor, you could open
      • “There’s a recent AI security experiment that sounds like science‑fiction, so I used a Battlestar Galactica metaphor to describe it